projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 20cb820) | patch
Use one task per iteration in parallel_for_loop.
authorRafael Espindola <rafael.espindola@gmail.com>
Tue, 15 Nov 2016 22:13:16 +0000 (22:13 +0000)
committerRafael Espindola <rafael.espindola@gmail.com>
Tue, 15 Nov 2016 22:13:16 +0000 (22:13 +0000)
commite5cd5ecd2ec73737cc7a38daeb7a38a4ad165d7a
treeff4170b6955892f4416aca3200099129f615104ftree | snapshot
parent20cb82088045cfa67ae4874e660287aac56042a7commit | diff
Use one task per iteration in parallel_for_loop.

This seems far more natural. A user can create larger chunks if the
overhead is too large.

With this linking xul with "--threads --build-id=sha1 goes from
13.938177535 to 11.035953538 seconds on linux.

llvm-svn: 287042
lld/include/lld/Core/Parallel.h diff | blob | history
Domain: System / Toolchain;